Research On Collision Avoidance Method Based On Braking Behavior

With the growing concern for the traffic safety, passive safety devices(seat belt, airbag, et al.) have been frequently criticized because of their drawbacks. Only concerning about the injuries of people from traffic accidents no longer meets the need for life safety and property protection. However, using the on-board radar, high speed video capture, ultrasonic detector and other advanced sensors to catch surrounding environmental information and information of vehicle per se is the new key means to solve traffic safety issues. Driving assistant system inevitably interferes the normal driving, even leads to driver’s operation mistakes. Therefore, it is useful to improve the driving assistant system through accurately knowing about the behavior information of drivers.Since the braking operation is a very common driving behavior, and the driver’s braking behavior is part of the driving. This paper studies the driver’s braking behavior and puts a new front obstacles warning strategy. Firstly, this paper concludes the factors which may affect the driver’s braking behavior through the analysis of the research on braking behavior both home and abroad. Then, calculate the contribution rates of all the factors which may affect driver’s behavior using the method of PCA(Principal Component Analysis), and establishes the driver’s braking behavior prediction model with BP neural network. Finally, the front obstacle alerting strategy will be built based on the braking deceleration as an alerting parameter. Main research contents of this paper are as follows:(1)Data collection of braking behavior. This paper designs an urban road driving scenes based on the dynamic steering automobile simulative instrument. Then the driving parameter data of the car-following has collected through recruiting the experimental drivers, which lay a solid foundation for characteristic parameter selection and model establishment.(2)Filtering of the characteristic parameters. Based on the research and the analysis of the factors which affect driver’s driving behaviors, 7 factors have been achieved and their contributions have also been calculated with the PCA theory. Finally, acceleration of ahead vehicle, relative speed, relative distance for both vehicles and reciprocal of collision time(TTCi) 4 parameters are determined can be inputted in the BP neural network model.(3)Building the driver’s braking behavior model based on BP neural network. The driver’s braking behavior model based on BP neural network is trained by inputting the data of the driver’s braking behavior, the accurate rate and error range are the evaluation standards. The BP neural network model’s nodes amount of enter layer-middle layer-output layer, threshold, learning rate and training function have been determined. And finally the driver’s braking behavior model is built.(4)The division of the braking intentions of drivers. After completing the braking behavior prediction of drivers, braking forecast data of each experimental driver is divided by introducing method of K-means analysis algorithm. Each experimenter’s braking forecast data is the initial clustering data, and 3 is set as the goal clustering barycenter. And then the overlapping part is further divided through the median method. Finally, three braking intentions of drivers can be classified: urgent braking, medium braking and slight braking.(5)Establishing the new front obstacle alerting strategy. Based on the speed, acceleration and relative distance of two vehicles and driver’s braking time delay, this paper proposed putting deceleration as an alerting parameter for alerting strategy. This strategy is verified by the steering automobile simulative instrument.(6)Evaluation of active collision avoidance strategies. Recollecting the experimental data of number(3) and(6), the front obstacle alerting value is calculated and the braking behaviors are predicted using the built braking behavior prediction model. At last, two alerting strategies are evaluated by using the standards of false alarm rate and the situation of warning.